sup-1-none-true-primal
one-step supervised training
sup-1-none-true-full_gradient
full gradient
sup-2-none-true-primal
two-step supervised training with aggregation over time
sup-2-none-true-full_gradient
full gradient
sup-2-none-true-no_net_bptt
no backpropagation through time over the network
sup-2-none-false-primal
two-step supervised training with loss only at final state
sup-2-none-false-full_gradient
full gradient
sup-2-none-false-no_net_bptt
no backpropagation through time over the network
sup-3-none-true-primal
three-step supervised training with aggregation over time
sup-3-none-true-full_gradient
full gradient
sup-3-none-true-no_net_bptt
no backpropagation through time over the network
sup-3-none-false-primal
three-step supervised training with loss only at final state
sup-3-none-false-full_gradient
full gradient
sup-3-none-false-no_net_bptt
no backpropagation through time over the network
sup-4-none-true-primal
four-step supervised training with aggregation over time
sup-4-none-true-full_gradient
full gradient
sup-4-none-true-no_net_bptt
no backpropagation through time over the network
sup-4-none-true-cut_every_2_net_bptt
stop backpropagation through time over the network every two network calls
mix-1-1-true-primal
mixed-chain with one network followed by one physics step with aggregation over time
mix-1-1-true-full_gradient
full gradient
mix-1-1-false-primal
mixed-chain with one network followed by one physics step with loss only at final state
mix-1-1-false-full_gradient
full gradient
mix-2-1-true-primal
mixed-chain with two network followed by one physics step with aggregation over time
mix-2-1-true-full_gradient
full gradient
mix-2-1-true-no_net_bptt
no backpropagation through time over the network
mix-2-1-false-primal
mixed-chain with two network followed by one physics step with loss only at final state
mix-2-1-false-full_gradient
full gradient
mix-2-1-false-no_net_bptt
no backpropagation through time over the network
div-2-1-true-primal
diverted-chain with two steps in main chain and one branch length
div-2-1-true-full_gradient
full gradient
div-2-1-true-no_dp
no differentiable physics
div-2-1-true-no_net_bptt
no backpropagation through time over the network
div-2-1-true-no_dp-no_net_bptt
no differentiable physics and no backpropagation through time over the network
div-3-1-true-primal
diverted chain with three steps in main chain and one branch step length
div-3-1-true-full_gradient
full gradient
div-4-1-true-primal
diverted chain with four steps in main chain and one branch step length
div-4-1-true-full_gradient
full gradient
div-3-2-true-primal
diverted chain with three steps in main chain and branch chain of length two
div-3-2-true-full_gradient
full gradient
div-3-2-false-primal
diverted chain with three steps in main chain and branch chain of length two with loss only at final state of the branch
div-3-2-false-full_gradient
full gradient
div-4-2-true-primal
diverted chain with four steps in main chain and branch chain of length two
div-4-2-true-full_gradient
full gradient
div-4-2-false-primal
diverted chain with four steps in main chain and branch chain of length two with loss only at final state of the branch
div-4-2-false-full_gradient
full gradient
tf-3-1-true-primal
three-step supervised rollout with state reset after each prediction
tf-3-1-true-full_gradient
full gradient
tf-4-2-true-primal
four-step supervised rollout with state reset after every second prediction
tf-4-2-true-full_gradient
full gradient
tf-4-2-true-no_net_bptt
no backpropagation through time over the network
tf-4-2-false-primal
four-step supervised rollout with state reset after every second prediction; loss only at end of each forcing period
tf-4-2-false-full_gradient
full gradient
tf-4-2-false-no_net_bptt
no backpropagation through time over the network
res-1-none-false-primal
one-step residuum training
res-1-none-false-full_gradient
full gradient
res-2-none-true-primal
two-step residuum training with aggregation over time
res-2-none-true-full_gradient
full gradient
res-2-none-true-no_net_bptt
no backpropagation through time over the network
res-3-none-true-primal
three-step residuum training with aggregation over time
res-3-none-true-full_gradient
full gradient
res-3-none-true-no_net_bptt
no backpropagation through time over the network
